1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
|
"""Nyströmer encoder.
Stolen from:
https://github.com/lucidrains/nystrom-attention/blob/main/nystrom_attention/nystrom_attention.py
"""
from typing import Optional
from torch import nn, Tensor
from text_recognizer.networks.transformer.mlp import FeedForward
from text_recognizer.networks.transformer.norm import PreNorm
from text_recognizer.networks.transformer.nystromer.attention import NystromAttention
class Nystromer(nn.Module):
def __init__(
self,
*,
dim: int,
depth: int,
dim_head: int = 64,
num_heads: int = 8,
num_landmarks: int = 256,
inverse_iter: int = 6,
residual: bool = True,
residual_conv_kernel: int = 33,
dropout_rate: float = 0.0,
glu: bool = True,
) -> None:
super().__init__()
self.dim = dim
self.layers = nn.ModuleList(
[
nn.ModuleList(
[
PreNorm(
dim,
NystromAttention(
dim=dim,
dim_head=dim_head,
num_heads=num_heads,
num_landmarks=num_landmarks,
inverse_iter=inverse_iter,
residual=residual,
residual_conv_kernel=residual_conv_kernel,
dropout_rate=dropout_rate,
),
),
PreNorm(
dim,
FeedForward(dim=dim, glu=glu, dropout_rate=dropout_rate),
),
]
)
for _ in range(depth)
]
)
def forward(self, x: Tensor, mask: Optional[Tensor] = None) -> Tensor:
for attn, ff in self.layers:
x = attn(x, mask=mask) + x
x = ff(x) + x
return x
|
